A multivariate analysis of the survival of patients with aggressive lymphoma - Variations in the predictive value of prognostic factors during the course of the disease

BACKGROUND. Aggressive non-Hodgkin's lymphoma is now often curable with chemotherapy. The International Prognostic Index (IPI) was recently developed to predict patient survival on the basis of pretreatment clinical features. However, classical multivariate regression models such as the IPI fail to detect time-dependent changes in the predictive value of covariates. In this study, an extension of the Cox proportional hazards model was used to determine whether the value of prognostic factors might decay over time. METHODS. A total of 1271 patients younger than 60 years, entered on the LNH-84 and LNH-87 studies of an ACVBP induction regimen (consisting of doxorubicin, cyclophosphamide, vindesine, bleomycin, methylprednisolone, and IT methotrexate), were analyzed in terms of overall survival and monthly risk of dying. By a standard method, prognostic factors were identified in a training sample and confirmed in a validation sample. The independently significant covariates were then included in a step-function regression model (3-step) that permitted determination of their value in predicting the short term and long term survival of the entire population. RESULTS. During the entire follow-up period (median, 5.5 years), lactate dehydrogenase level, tumor stage, performance status, and number of extranodal sites remained independently predictive of overall survival. However, these covariates had nonproportional hazard functions. The study of their time-dependent effect with the 3-step model confirmed that they were predictive of overall survival during the short term follow-up period of 3 months to 2 years. However, during the induction period of 0-3 months and the long term follow-up period of 2-10 years, there was only 1 independently predictive factor for each of these periods: performance status and tumor stage, respectively. CONCLUSIONS. The IPI factors are relevant to short term follow-up and permit the selection of routine or experimental therapeutic regimens. In contrast, only performance status is predictive of a patient's ability to tolerate induction chemotherapy, and only tumor stage is predictive of long term survival. (C) 1998 American Cancer Society